Anki FSRS Explained: How the New Scheduler Works
Open any Anki forum today and FSRS is the topic: the machine-learning scheduler that replaced the thirty-year-old SM-2 algorithm and now ships enabled by default for new Anki profiles. This guide explains what it actually does, in plain English, and walks you through switching an existing collection without drowning in a backlog. If you are new to spaced repetition itself, start with our spaced repetition guide and come back.
What FSRS Changes, Compared to SM-2
In an open benchmark on tens of thousands of real Anki collections, FSRS predicts recall more accurately than SM-2 and other classic schedulers, which is why it became Anki's default for new profiles. Open Spaced Repetition benchmark (GitHub)
FSRS stands for Free Spaced Repetition Scheduler. It answers the same question SM-2 did, when should this card come back, but it answers it with a statistical model of your memory instead of a fixed formula from 1987. Four differences explain nearly all of the hype:
Difficulty, Stability, Retrievability
FSRS builds on the DSR memory model from research on optimizing spaced repetition schedules, published at KDD 2022 and trained on hundreds of millions of real-world reviews. Ye et al., 2022 (KDD)
Every card in an FSRS collection carries three numbers, and every review updates them. Understanding these three is understanding FSRS:
An irregular verb you keep mixing up carries high difficulty, so its intervals grow cautiously even on a good streak.
A card with a stability of 60 days can safely wait two months before your recall odds dip below nine in ten.
When a card's predicted recall slides to 90% and your target is 90%, that card shows up in today's queue.
Desired Retention: The One Setting That Matters
Under SM-2 you tuned half a dozen knobs and the retention you got was whatever fell out. FSRS collapses all of that into a single number: desired retention, the probability of recall you want a card to have at the moment it comes due. Set it to 0.90 and every card is scheduled to reappear exactly when the model predicts your chance of remembering it has slipped to 90%.
The trade-off is workload. Retention and daily reviews do not scale together linearly: moving from 0.90 to 0.95 costs noticeably more reviews per day, and Anki warns outright once you push past 0.97. In the other direction, dropping to 0.85 or 0.80 thins the queue dramatically while you still keep most of the material.
The practical advice: start at the default 0.90, live with it for a few weeks, then adjust by feel. Overwhelmed by reviews? Lower it. Preparing for an exam where every card counts? Raise it for that deck's preset only. Being able to make that decision per deck, with one honest number, is the quality-of-life upgrade FSRS users talk about most.
Switching to FSRS, Step by Step
Update Anki
- If you started your collection recently on an up-to-date Anki, you may already be on FSRS: open any deck's options and look for the FSRS section.
- Existing collections are never switched automatically. If you have been using Anki for years, you are still on SM-2 until you flip the toggle yourself.
- Update AnkiDroid and AnkiMobile too. All current clients understand FSRS scheduling, and mixing a modern desktop with a very old mobile client is the one setup that causes trouble.
Flip the FSRS toggle
- Your review history is untouched. FSRS reads the same log SM-2 wrote, so years of reviews immediately inform the new schedule.
- SM-2 knobs like ease bonus, interval modifier, and graduating interval disappear from the options screen, because FSRS replaces all of them.
- There is no lock-in: you can switch back to SM-2 later, though after a few weeks on FSRS almost nobody does.
Set desired retention
- The default of 0.90 is a sensible balance of retention and workload for most learners.
- Anki itself warns against going above 0.97: the closer you push toward certainty, the faster the daily review count explodes.
- You can give different presets different targets: 0.95 for an exam deck, 0.85 for casual vocabulary you just want to keep alive.
Optimize the parameters
- With little history the optimizer has nothing to learn from. The defaults were trained on an enormous corpus of real reviews, so starting with them is fine.
- Once you have a few hundred reviews per preset, optimizing starts to personalize the schedule meaningfully.
- FSRS-6 even fits the shape of your personal forgetting curve, which is one of the reasons it benchmarks ahead of earlier versions.
Leave rescheduling off for now
- Rescheduling a mature collection typically dumps a large backlog of suddenly-due cards on you, which is the single most common reason people rage-quit the switch.
- Left off, cards move onto FSRS scheduling naturally, one review at a time, and the transition is invisible.
- You can always reschedule later, deliberately, once you understand the workload it creates.
Simplify your learning steps
- Keep one or two short steps, something like 1m 10m, so a brand-new card gets a same-session check before FSRS schedules it.
- Keep every step under one day. Repeating a card many times within a single day adds little long-term memory, and steps of a day or more collide with the intervals FSRS wants to set.
- Delete elaborate multi-day step schemes left over from SM-2 guides. They were workarounds for a scheduler you are no longer using.
Mistakes That Break the Model
FSRS is a model of your memory, so it is only as good as the answers you feed it. These six habits account for almost every "FSRS does not work for me" thread.
Hard is a passing grade. FSRS reads it as 'recalled, with effort' and grows the interval. Using it as a soft fail teaches the model your memory is stronger than it is, inflating every future interval. If the answer did not come back, the button is Again, full stop.
Workload does not scale linearly with retention. Each point above roughly 0.95 buys a sliver of recall for a disproportionate pile of extra reviews, which is why Anki warns you above 0.97. If reviews feel crushing, lowering desired retention is the intended fix.
FSRS parameters describe one person's memory. Pasting someone else's set gives you a schedule fitted to their brain and their deck. Use the defaults or your own Optimize run, never borrowed numbers.
Parameters shift slowly, because one week of reviews barely changes a history of thousands. Re-optimizing about once a month is plenty; more often just burns time for identical numbers.
Switching to FSRS does not reset anything. The model reads your existing history and picks up where SM-2 left off, so resetting cards to relearn them throws away exactly the data FSRS is best at using.
Seeing a two-month jump on a card you answered twice feels wrong if you are used to SM-2's cautious ladder. But the interval is a prediction with your target retention built in. Give it a few weeks and check your true retention statistics before overriding anything.
Fine-Tuning: Optimization Cadence, Parameters, Easy Days
The defaults carry you a long way. These are the adjustments worth making once you have lived with FSRS for a month and want the schedule to fit your life, not the other way around.
Optimize roughly once a month per preset, and additionally after any big change in how you study: a new deck type, a long break, or a switch in what the deck tests. Parameters live per preset, so decks that behave differently (recognition vs production, formulas vs vocabulary) benefit from separate presets with separate optimization runs.
The first four numbers are the initial stability granted by your first grade on a card (Again, Hard, Good, Easy); the rest shape how difficulty and stability evolve and, in FSRS-6, the curvature of your forgetting curve. They are diagnostics, not dials: resist hand-editing them, because a single miscopied value can distort every schedule in the preset.
Desired retention is where you express priorities. A medical-exam preset at 0.93 to 0.95 accepts the heavy review load; a leisure-vocabulary preset at 0.83 to 0.87 trades a few forgotten words for a much lighter daily queue. Changing the number only affects future scheduling unless you also reschedule, so experiments are cheap.
Anki's Easy Days feature lets you mark weekdays when you want a lighter queue, and the scheduler nudges due dates around them. Because FSRS knows each card's predicted recall, it can shift reviews a day or two at minimal retention cost, which beats the old approach of letting a backlog pile up every weekend.
Turn it on deliberately in exactly two situations: you have just optimized after months of drift and accept a one-time backlog to align everything, or you changed desired retention and want the whole deck moved to the new target now. Do it on a day with spare time, clear the backlog, and turn it back off.
Premium access includes:
FAQ
Anki FSRS: common questions
Quick answers to the questions learners ask most when switching to FSRS.
Is FSRS really better than SM-2?
For almost everyone, yes. In open benchmarks on thousands of real Anki collections, FSRS predicts recall more accurately than SM-2, and a more accurate prediction means reviews land closer to the ideal moment. In practice that shows up as the same retention with noticeably fewer daily reviews. It is why Anki now enables FSRS by default for new profiles.
Will switching to FSRS mess up my existing decks?
No. FSRS reads the review history you already have and schedules from it; nothing is deleted or reset. If you leave 'Reschedule cards on change' off, existing cards simply move onto FSRS scheduling one review at a time and you will barely notice the transition.
What desired retention should I pick?
Start with the default 0.90. Drop toward 0.85 if the daily queue feels heavy and the material is low-stakes; raise it toward 0.95 for exam-critical decks and accept the extra reviews. Anki warns against going above 0.97 because workload grows steeply near certainty.
How often should I optimize FSRS parameters?
About once a month is enough. Parameters are fitted to your whole review history, so a few extra weeks of data moves them very little. Optimize sooner only after a major change, like returning from a long break or adding a very different kind of deck.
Should I still press Hard, and what about learning steps?
Press Hard only when you recalled the answer with effort; if you forgot, press Again. Keep learning steps short and few, for example 1m 10m, and never a day or longer, so they do not fight the intervals FSRS computes.
Do decks generated with NextLang work with FSRS?
Yes. NextLang exports standard .apkg files, and FSRS schedules any card in your collection regardless of where it came from. Generate a deck, import it, and FSRS starts modeling each card from your very first review.
Give FSRS Something Worth Scheduling
Generate a complete Anki deck with translations and example sentences, import the .apkg, and let FSRS handle every review from day one.
Keep reading
Related guides that build on what you just read.